Call Center Multimodal Voice Search Matt Yuschik, Ph.D. Human Factors Specialist, Human Computer Interactions Relationship Technology Management Session A3 Voice User Interface in Voice Search Voice Search Conference March 2-4, 9
Convergys Corporation A Global Leader in Relationship Management Worldwide Capabilities 75, employees 85 contact centers, and clients in 7+ countries A Leading Public Company $2.8 billion in revenues Listed on NYSE, S&P, Fortune A Fortune Most Admired Company for eight consecutive years Key Facts About Convergys Host more than 1 billion customer interactions annually Support more than 3 million employees and retirees worldwide Billing for 3+ million communications subscribers worldwide Top for Innovative Use of Technology of InformationWeek (7) 8 Convergys Corporation. All rights reserved. 2
Analytics Analytics Analytics Analytics Analytics Relationship Technology - Call Center Solutions Customer Needs (Per Forrester) Consistent Channel Experience Advanced Self-Service Capabilities Live e-channels Support Agents Who Know My History Proactive Notification and Support Extended Hours of Service Multi-Channel Self-Care Speech Solution Outside-In Approach to match customer needs to client needs Customer Data and Interactions Hub Client Needs (Per Yankee Group) Enhanced Customer Experience Improved Revenue Opportunities Lower Operational Costs Automated Self-Care Virtual/Secret Agents (Self-Care Optimizer) Real-time Decisioning Web Solution Mobile Device Solution Dynamic Decisioning Solution 8 Convergys Corporation. All rights reserved. 3
Path from Multimodal Agent to End-User Self-care Point Solutions vs. Solution Path 7 k SMEs Solutioners Observers MM testers Transactions Dialogs Tasks Subtasks Multimodal presentation Conversational Interface Speaker Verification for security Hidden Agent for difficulties Device-Independent Multichannel 8 Convergys Corporation. All rights reserved. 4
Multimodal Preferences What devices and combination of modes are available? What tasks can be done with multimodality, and which should? device landline cellphone PC PDA 3G phone Input speak x x x x x type x x x x tap x x GPS x Output listen x x x x x listen - TTS x read text x x x x view figures x x x view video x x Input: What should be spoken? What should be typed or pressed? Output: What should be read? What should be heard? 8 Convergys Corporation. All rights reserved. 5
Needs Assessment and Solution- Delivery Service Outside-In Approach = Monitor Agents and Callers Start with GUI based Service Handling Many layers / screens to complete a call Other 32% Sevice Types Complaints 12% Address 5% Rates 5% Monitor Services > Transaction types / Use Cases Hold 3% Identify most frequent functions ZIP code 7% Redeliver 4% Decompose into Tasks Navigation, Form Filling, Data Caching, Search Hours and Locations 11% Call Distribution Track 21% Solution = A Multimodal Agent Tool that includes Voice Search Caller value-add Navigation, Data-driven search Better Customer Care Experience Agent value-add Improves Agent Productivity and Quality Better Agent Satisfaction, increased retention 8 Convergys Corporation. All rights reserved. 6
sec sec sec Simulation Results - Longitudinal Longitudinal testing Gives agents time to learn Identifies time training Test Conditions 4 Agents (6 mo - yrs experience) 1 UI per day for 3 days 5 Services 7 repetitions / svc expect leveling No caller involved UI Types wrapper concept Existing GUI Narrow voice activate existing flow (AA) Broad streamline, follow agent/caller dialog (A2) 7 6 trial 7 6 trial E1- H ol d A H T 1 2 3 4 5 6 7 A A - H ol d A H T 1 2 3 4 5 6 7 8 Convergys Corporation. All rights reserved. 7 7 6 trial A2 - Hold AHT 1 2 3 4 5 6 7
sec sec sec sec Pilot Test Learning Types 6 - - - - 6 - - - - 6 - - - - Level 1 Agents 1 2 3 4 5 6 7 8 9 11 12 13 14 15 16 17 18 19 day Level 2 Agents 1 2 3 4 5 6 7 8 9 11 12 13 14 15 16 17 18 19 days Level 3 Agents 1 2 3 4 5 6 7 8 9 11 12 13 14 15 16 17 18 19 day 8 Convergys Corporation. All rights reserved. 8 Excellent Learners get it quickly <7 days Normal Learners accomodate then decrease -15 days Delayed Learners At risk Don t want to get it 7 6 - - Delta AHT for Level 2 Agents (without E1 bias) 1 2 3 4 5 6 7 8 9 11 12 13 14 15 16 17 day Expected Results
Change from Baseline Multimodal AHT- Performance Groups AVA Group - w/o CTI Agents - - - - 1 2 3 4 5 6 7 8 9 11 12 13 14 15 Days on AVA 8 Convergys Corporation. All rights reserved. 9 Faster(n = 7) Little Change (n = 6) Slower (n = 5) Successful Agent Mismatched Agent Comfortable with new technology Shorter tenure Uses most of the MMUI most of the time Uses conference mode Longer tenure Resistant to technology changes Extensive use of Mute mode Dependent on a notepad or scratchpad
Multimodal UI Design Principles Voice Search Speech and graphics are both active at the same time GUI and VUI are integrated, and do not just work in parallel Error conditions are handled in both modalities Shortcuts for common transactions (mixed initiative dialog) Multi-word utterances (data driven phrases) Timeout when default step is next time to accept/reject Limited acknowledgement The agent and the system are focused on one vocabulary, the same vocabulary Don t need SLMs or NL for the most part (constrained domain) Provide a temporary memory for data and speech 8 Convergys Corporation. All rights reserved.
Conclusions and Next Steps Wrapper requirements and effects API approach Component Applications Voice Pad Evolution to Multi-Modal Self-care devices Application Assessment Process Determining best places to add another modality (voice) Categories of Transactions Customer service hot-spots Categorizing flows and tasks Training What should be learned, and when Useful tools and practices 8 Convergys Corporation. All rights reserved. 11
Questions, Comments, Insights? Matt Yuschik, Ph.D. Human Factors Specialist, Human Computer Interactions Relationship Technology Management Multi-Channel Self-Care Solutions Convergys Corporation matt.yuschik@convergys.com 12